huff0: Pass a single bitReader pointer to asm #634
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
klauspost
approved these changes
Jun 30, 2022
|
Removed the /+1000/. I also found that the return value computation could be made one instruction shorter. I haven't rerun the benchmarks, since I don't expect that to have a significant effect. |
This makes the context object smaller and frees up three registers, which we can use to replace the limitPtr and bufferOrigin stack variables. Benchmark results show a tiny win (Go 1.19beta, Core i7-3770K): name old speed new speed delta Decompress1XTable/digits-8 347MB/s ± 0% 347MB/s ± 0% ~ (p=0.650 n=8+10) Decompress1XTable/gettysburg-8 268MB/s ± 0% 268MB/s ± 0% ~ (p=0.400 n=9+9) Decompress1XTable/twain-8 327MB/s ± 0% 327MB/s ± 1% ~ (p=0.339 n=7+9) Decompress1XTable/low-ent.10k-8 385MB/s ± 0% 385MB/s ± 1% ~ (p=0.510 n=9+10) Decompress1XTable/superlow-ent-10k-8 376MB/s ± 0% 376MB/s ± 0% ~ (p=0.712 n=8+10) Decompress1XTable/crash2-8 17.3MB/s ± 1% 17.3MB/s ± 1% ~ (p=0.926 n=10+10) Decompress1XTable/endzerobits-8 52.9MB/s ± 1% 52.4MB/s ± 0% -0.94% (p=0.000 n=10+10) Decompress1XTable/endnonzero-8 11.4MB/s ± 0% 11.4MB/s ± 1% ~ (p=0.343 n=10+10) Decompress1XTable/case1-8 22.0MB/s ± 0% 22.0MB/s ± 0% ~ (p=0.618 n=9+9) Decompress1XTable/case2-8 18.1MB/s ± 0% 18.1MB/s ± 0% ~ (p=0.348 n=9+9) Decompress1XTable/case3-8 19.1MB/s ± 0% 19.1MB/s ± 0% +0.21% (p=0.048 n=10+10) Decompress1XTable/pngdata.001-8 374MB/s ± 0% 374MB/s ± 0% ~ (p=0.861 n=9+10) Decompress1XTable/normcount2-8 54.3MB/s ± 1% 54.5MB/s ± 1% ~ (p=0.093 n=10+10) Decompress1XNoTable/digits/100-8 279MB/s ± 0% 280MB/s ± 0% +0.30% (p=0.003 n=10+9) Decompress1XNoTable/digits/10000-8 366MB/s ± 0% 365MB/s ± 0% ~ (p=0.113 n=10+9) Decompress1XNoTable/digits/262143-8 347MB/s ± 0% 347MB/s ± 1% ~ (p=0.739 n=10+10) Decompress1XNoTable/gettysburg/100-8 278MB/s ± 1% 277MB/s ± 1% ~ (p=0.676 n=10+9) Decompress1XNoTable/gettysburg/10000-8 363MB/s ± 1% 362MB/s ± 0% -0.50% (p=0.001 n=10+9) Decompress1XNoTable/gettysburg/262143-8 350MB/s ± 0% 347MB/s ± 0% -0.90% (p=0.000 n=10+8) Decompress1XNoTable/twain/100-8 268MB/s ± 0% 267MB/s ± 0% ~ (p=0.384 n=9+8) Decompress1XNoTable/twain/10000-8 363MB/s ± 0% 362MB/s ± 0% -0.32% (p=0.000 n=9+9) Decompress1XNoTable/twain/262143-8 328MB/s ± 0% 329MB/s ± 0% ~ (p=0.063 n=9+10) Decompress1XNoTable/low-ent.10k/100-8 180MB/s ± 0% 181MB/s ± 0% ~ (p=0.225 n=10+10) Decompress1XNoTable/low-ent.10k/10000-8 385MB/s ± 0% 385MB/s ± 0% ~ (p=0.289 n=10+10) Decompress1XNoTable/low-ent.10k/262143-8 389MB/s ± 1% 389MB/s ± 1% ~ (p=0.971 n=10+10) Decompress1XNoTable/superlow-ent-10k/262143-8 389MB/s ± 0% 390MB/s ± 0% +0.27% (p=0.017 n=9+10) Decompress1XNoTable/crash2/100-8 278MB/s ± 0% 279MB/s ± 1% ~ (p=0.163 n=9+10) Decompress1XNoTable/crash2/10000-8 373MB/s ± 1% 373MB/s ± 0% ~ (p=0.370 n=10+8) Decompress1XNoTable/crash2/262143-8 375MB/s ± 0% 375MB/s ± 0% ~ (p=0.604 n=9+10) Decompress1XNoTable/endzerobits/100-8 180MB/s ± 0% 181MB/s ± 0% +0.26% (p=0.005 n=10+9) Decompress1XNoTable/endzerobits/10000-8 384MB/s ± 0% 385MB/s ± 0% ~ (p=0.914 n=8+10) Decompress1XNoTable/endzerobits/262143-8 389MB/s ± 0% 390MB/s ± 0% ~ (p=0.739 n=10+10) Decompress1XNoTable/endnonzero/100-8 180MB/s ± 1% 180MB/s ± 1% ~ (p=0.926 n=10+10) Decompress1XNoTable/endnonzero/10000-8 384MB/s ± 0% 384MB/s ± 0% ~ (p=0.965 n=10+8) Decompress1XNoTable/endnonzero/262143-8 390MB/s ± 0% 390MB/s ± 0% ~ (p=0.633 n=8+10) Decompress1XNoTable/case1/100-8 282MB/s ± 0% 283MB/s ± 0% +0.34% (p=0.005 n=10+10) Decompress1XNoTable/case1/10000-8 372MB/s ± 0% 373MB/s ± 0% ~ (p=0.113 n=9+9) Decompress1XNoTable/case1/262143-8 374MB/s ± 0% 374MB/s ± 0% ~ (p=0.448 n=10+10) Decompress1XNoTable/case2/100-8 274MB/s ± 1% 274MB/s ± 0% ~ (p=0.927 n=10+10) Decompress1XNoTable/case2/10000-8 376MB/s ± 0% 376MB/s ± 0% ~ (p=0.408 n=10+8) Decompress1XNoTable/case2/262143-8 376MB/s ± 1% 377MB/s ± 0% ~ (p=1.000 n=10+10) Decompress1XNoTable/case3/100-8 266MB/s ± 0% 265MB/s ± 0% ~ (p=0.113 n=9+10) Decompress1XNoTable/case3/10000-8 372MB/s ± 0% 372MB/s ± 0% ~ (p=0.075 n=10+9) Decompress1XNoTable/case3/262143-8 374MB/s ± 0% 374MB/s ± 0% ~ (p=0.172 n=10+10) Decompress1XNoTable/pngdata.001/100-8 238MB/s ± 0% 238MB/s ± 0% ~ (p=0.438 n=9+8) Decompress1XNoTable/pngdata.001/10000-8 384MB/s ± 0% 384MB/s ± 0% ~ (p=0.448 n=10+10) Decompress1XNoTable/pngdata.001/262143-8 378MB/s ± 0% 378MB/s ± 0% ~ (p=0.836 n=10+10) Decompress1XNoTable/normcount2/100-8 281MB/s ± 0% 282MB/s ± 1% ~ (p=0.122 n=8+10) Decompress1XNoTable/normcount2/10000-8 369MB/s ± 1% 369MB/s ± 0% ~ (p=0.912 n=10+10) Decompress1XNoTable/normcount2/262143-8 370MB/s ± 0% 370MB/s ± 1% ~ (p=0.342 n=10+10) Decompress4XNoTable/digits/100-8 197MB/s ± 0% 197MB/s ± 1% ~ (p=0.764 n=10+9) Decompress4XNoTable/digits/10000-8 594MB/s ± 0% 602MB/s ± 1% +1.35% (p=0.000 n=10+10) Decompress4XNoTable/digits/262143-8 570MB/s ± 1% 578MB/s ± 0% +1.30% (p=0.000 n=10+8) Decompress4XNoTable/gettysburg/100-8 258MB/s ± 1% 260MB/s ± 0% +0.59% (p=0.001 n=10+10) Decompress4XNoTable/gettysburg/10000-8 638MB/s ± 0% 641MB/s ± 0% +0.44% (p=0.000 n=9+9) Decompress4XNoTable/gettysburg/262143-8 573MB/s ± 1% 574MB/s ± 0% ~ (p=0.353 n=10+10) Decompress4XNoTable/twain/100-8 214MB/s ± 2% 214MB/s ± 2% ~ (p=0.853 n=10+10) Decompress4XNoTable/twain/10000-8 634MB/s ± 1% 638MB/s ± 0% +0.62% (p=0.000 n=10+10) Decompress4XNoTable/twain/262143-8 513MB/s ± 1% 517MB/s ± 0% +0.85% (p=0.000 n=10+10) Decompress4XNoTable/low-ent.10k/100-8 195MB/s ± 0% 194MB/s ± 0% ~ (p=0.130 n=9+9) Decompress4XNoTable/low-ent.10k/10000-8 635MB/s ± 0% 642MB/s ± 0% +1.19% (p=0.000 n=10+10) Decompress4XNoTable/low-ent.10k/262143-8 675MB/s ± 0% 685MB/s ± 0% +1.51% (p=0.000 n=10+10) Decompress4XNoTable/superlow-ent-10k/262143-8 673MB/s ± 1% 684MB/s ± 0% +1.70% (p=0.000 n=10+10) Decompress4XNoTable/case1/100-8 206MB/s ± 1% 206MB/s ± 0% ~ (p=0.189 n=10+9) Decompress4XNoTable/case1/10000-8 593MB/s ± 0% 601MB/s ± 0% +1.47% (p=0.000 n=10+10) Decompress4XNoTable/case1/262143-8 603MB/s ± 0% 613MB/s ± 0% +1.64% (p=0.000 n=10+10) Decompress4XNoTable/case2/100-8 201MB/s ± 0% 202MB/s ± 1% ~ (p=0.053 n=9+10) Decompress4XNoTable/case2/10000-8 610MB/s ± 0% 618MB/s ± 0% +1.30% (p=0.000 n=9+10) Decompress4XNoTable/case2/262143-8 622MB/s ± 1% 634MB/s ± 0% +1.90% (p=0.000 n=9+8) Decompress4XNoTable/case3/100-8 197MB/s ± 1% 198MB/s ± 0% +0.53% (p=0.001 n=9+10) Decompress4XNoTable/case3/10000-8 606MB/s ± 0% 615MB/s ± 0% +1.49% (p=0.000 n=8+10) Decompress4XNoTable/case3/262143-8 613MB/s ± 1% 622MB/s ± 0% +1.48% (p=0.000 n=10+10) Decompress4XNoTable/pngdata.001/100-8 212MB/s ± 1% 211MB/s ± 0% ~ (p=0.136 n=9+9) Decompress4XNoTable/pngdata.001/10000-8 645MB/s ± 1% 649MB/s ± 1% +0.65% (p=0.000 n=9+10) Decompress4XNoTable/pngdata.001/262143-8 640MB/s ± 1% 649MB/s ± 0% +1.44% (p=0.000 n=10+10) Decompress4XNoTable/normcount2/100-8 260MB/s ± 1% 261MB/s ± 1% ~ (p=0.211 n=10+9) Decompress4XNoTable/normcount2/10000-8 584MB/s ± 1% 591MB/s ± 0% +1.33% (p=0.000 n=9+9) Decompress4XNoTable/normcount2/262143-8 588MB/s ± 1% 596MB/s ± 1% +1.39% (p=0.000 n=10+9) Decompress4XNoTableTableLog8/digits-8 583MB/s ± 1% 592MB/s ± 0% +1.48% (p=0.000 n=10+10) Decompress4XTable/digits-8 580MB/s ± 0% 588MB/s ± 0% +1.33% (p=0.000 n=8+10) Decompress4XTable/gettysburg-8 368MB/s ± 1% 370MB/s ± 0% +0.59% (p=0.017 n=10+9) Decompress4XTable/twain-8 510MB/s ± 0% 515MB/s ± 0% +0.99% (p=0.000 n=9+10) Decompress4XTable/low-ent.10k-8 657MB/s ± 0% 665MB/s ± 0% +1.24% (p=0.000 n=10+10) Decompress4XTable/superlow-ent-10k-8 608MB/s ± 0% 617MB/s ± 1% +1.48% (p=0.000 n=8+10) Decompress4XTable/case1-8 21.1MB/s ± 1% 21.0MB/s ± 2% ~ (p=0.223 n=10+10) Decompress4XTable/case2-8 17.6MB/s ± 0% 17.6MB/s ± 0% ~ (p=0.199 n=9+10) Decompress4XTable/case3-8 18.7MB/s ± 0% 18.7MB/s ± 0% ~ (p=0.557 n=10+8) Decompress4XTable/pngdata.001-8 633MB/s ± 1% 645MB/s ± 0% +1.90% (p=0.000 n=9+10) Decompress4XTable/normcount2-8 49.9MB/s ± 1% 49.5MB/s ± 1% -0.64% (p=0.002 n=10+10) [Geo mean] 270MB/s 271MB/s +0.36%
|
Thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This makes the context object smaller and frees up three registers, which we can use to replace the limitPtr and bufferOrigin stack
variables.
Benchmark results show a tiny win (Go 1.19beta, Core i7-3770K):